CENTRE FOR APPLIED MACRO – AND PETROLEUM ECONOMICS (CAMP)
CAMP Working Paper Series No 9/2016
Mending the broken link:
heterogeneous bank lending and monetary policy pass-through.
Carlo Altavilla, Fabio Canova and Matteo Ciccarelli
© Authors 2016
This paper can be downloaded without charge from the CAMP website http://www.bi.no/camp
Mending the broken link: heterogeneous bank lending and monetary policy pass-through.
Carlo Altavilla, ECB
Fabio Canova,
Norwegian Business School y
Matteo Ciccarelli, ECBz October 31, 2016
Abstract
We analyze the pass-through of monetary policy measures to lending rates to …rms and households in the euro area using a novel bank-level dataset. Banks’ characteristics such as the capital ratio, the exposure to sovereign debt, and the percentage of non-performing loans are responsible for the heterogeneity in pass-through of conventional monetary pol- icy changes. The location of a bank is irrelevant. Non-standard measures normalized the capacity of banks to grant loans. Banks with high level of non-performing loans and low capital ratio were most a¤ected. Banks’lending margins fell considerably. Macroeconomic implications are discussed.
JEL Classi…cation numbers: C23, E44, E52, G21.
Keywords: Monetary policy pass-through, european banks, heterogeneity, VARs.
European Central Bank, Sonnemannstrasse 22,60314 Frankfurt, Germany; email:
carlo.altavilla@ecb.europa.eu
yCorresponding author, Norwegian Business School, CAMP, and CEPR. Department of Economics, BI Nor- wegian Business School, Nydalsveien 37, 0484 Oslo, Norway; email: fabio.canova@bi.no
zEuropean Central Bank, Sonnemannstrasse 22, 60314 Frankfurt, Germany; email mat- teo.ciccarelli@ecb.europa.eu.
1 Introduction
Lending conditions are crucial to the level of economic activity and welfare. This is particularly true in the euro area since bank loans are around 50% of the external balance sheet …nancing of both small and large non-…nancial corporations - in the U.S. bank loans account for only around 25%. If …rms face working capital and wage bill constraints, impairment of lending activities constrains hiring and investment decisions and, thus, the level of aggregate activity.
Lending conditions also matter for monetary policy: an ine¤ective pass-through of policy rate changes makes it much harder for a central bank to regulate the dynamics of aggregate demand. From the early 2000s to the end of 2007, monetary policy pass-through in the euro area was relatively homogeneous across countries (see e.g. Ciccarelli et al. 2013) and almost complete in the long run (see Hristov et al. 2014). During the …nancial and the sovereign debt crises, the situation changed. Figure 1 shows the policy rate and the distribution of lending rates to non-…nancial corporations for the collection of banks in our data set for the period July 2007 –December 2015, normalised so that the policy rate is set to zero in July 2007.
There are four distinct phases. Up to the end of 2008, the distribution of lending rates followed the dynamics of the policy rate and the spread of the cross-sectional distribution was small. From the beginning of 2009 to the middle of 2011, the median of the distribution of lending rates continued to follow policy rate changes, but the spread in the distribution increased.
In the third phase, the median of the lending rates distribution no longer follows policy rate changes, especially for banks operating in …nancially stressed countries – we term "stressed countries" Greece, Cyprus, Italy, Spain, Ireland and Portugal, while the "non-stressed countries"
are Austria, France, Germany, Netherlands, Belgium and Luxembourg. In this period, lending rate heterogeneity became considerable and di¤erent banks reacted di¤erently to monetary policy changes, even within each group of countries. For example, between September 2011 and May 2014, the range of the distribution of lending rates in non-stressed countries increased one percentage point, while the median fell from 3.2 to 2.3 percent. From May 2014 onwards, both the median and the spread of the lending rate distribution accompanied policy rate declines and the fall was considerable for stressed countries. These patterns beg obvious questions: Why did the pass-through change during the period of …nancial turmoil? Why did di¤erent banks respond di¤erently to monetary policy changes? Why was the pattern reversed after 2014?
Which feature of monetary policy was responsible for the change?
The conventional view in normal times is that balance sheet characteristics determine how banks react to monetary policy contractions. In the US, the pass-through to lending rates has
been found to be stronger for small (Kashyap and Stein, 1995), illiquid (Stein, 1998; Kashyap and Stein, 2000) and poorly capitalised banks (Peek and Rosengren, 1995; Kishan and Opiela, 2000; Van den Heuvel, 2002). Gambacorta (2005, 2008) con…rms these …ndings using a sample of Italian banks. Larger, better capitalised, and more liquid banks are more resilient to monetary contractions as they can more easily substitute sources of external …nancing, absorb expected future losses, and divert liquidity to satisfy increases in loan demand. Contractionary policy decisions are thus transmitted to the real economy mainly through small, illiquid, and poorly capitalised banks. In closed economies where …rms externally …nance a large portion of their operations through banks with a weak balance sheet position, monetary contractions are a powerful brake on domestic real activity.
In periods of …nancial stress, economic and regulatory constraints might alter the e¤ectiveness of monetary policy. In addition, and apart from two brief episodes, monetary policy in the euro area in the period of interest was expansionary and it is unclear as yet whether banks respond in the same way to expansionary and to contractionary policy changes. Recent work analysing the pass-through during the last decade in the euro area has produced contradictory conclusions. For example, Hristov et al. (2014), and Holton and Rodriguez d’Acri (2015) document a signi…cant fall in the average pass-through relative to the pre-crisis period, while Von Borstel et al. (2015), Illes et al. (2015) only …nd a mild decline which is statistically similar in core and periphery countries, once banks’e¤ective cost of funding is taken into account. Thus, variations in pass- through are attributed to area-wide structural changes. Jimenez et al. (2012), Acharya et al.
(2015), on the other hand, …nd that the health of banks’ balance sheet a¤ect their portfolio choices (see also Acharya and Ste¤en, 2015, Altavilla et al., 2016, and Acharya et al., 2016) and thus the transmission of monetary policy changes.
This paper re-examines the monetary pass-through to lending rates in the euro area during the turbulent 2007-2015 period using a novel monthly disaggregated data set. Contrary to previous work, which has used banks from one country only, a limited number of European banks, or country aggregates of bank data, our data set is very rich. We have information on the major euro area banks, covering about 75% of the total; we know their locations and a number of their balance sheet characteristics and this help us to avoid cross-sectional or cross country heterogeneity biases. Furthermore, the sample is su¢ ciently long to meaningfully distinguish between the pass-through of conventional and non-standard policy decisions, and to measure the contribution of the latter to normalising lending conditions.
We …rst examine how persistent policy rate surprises a¤ect the quartiles of the distribution of the lending rates to non-…nancial corporations for the period up to May 2014 - we call this
conventional policy pass-through. We show that, once country-speci…c factors are taken into account, the distinction between stressed and non-stressed countries is irrelevant and the median medium run pass-through for both types of countries is only around 0.6, much lower than in the early 2000s, and close to the estimates obtained using aggregated country data. Instead, for the period and for the collection of banks we consider, we …nd that bank balance sheet characteristics matter. In particular, the level of capitalisation and the exposure to sovereign debt are robust determinants of the spread of the distribution of pass-through. We estimate that the di¤erence between the top and the bottom quartiles of the distribution of lending rates pass-through sorted by these characteristics could be up to 50 basis points. The composition of a bank’s portfolio and its expected losses also help to explain the dynamics and the spread in the distribution of the monetary pass-through for large banks. Interestingly, these conclusions hold when we consider both lending rates to non-…nancial corporations and to households.
In the second part, we examine the pass-through of non-standard policies. Since June 2014, the ECB employed credit easing measures – the targeted longer-term re…nancing operations (TLTROs) - to "enhance the transmission of monetary policy and to reinforce the accommodative monetary policy stance in view of the (...) subdued monetary and credit dynamics" (ECB Economic Bulletin, October 2015). In January 2015, it also announced quantitative easing measures – the expanded asset purchase programme (APP) - to further ease monetary policy given that the policy rate had hit the zero lower bound and prospects for the real economy had deteriorated since the introduction of the credit easing package (see Altavilla et al., 2015).
We show that these measures helped to normalise lending conditions, reduce the cross- sectional dispersion of lending rates and produce a larger pass-through in the medium run.
Better lending conditions for non-…nancial corporations materialised because the instantaneous pass-through improved and because of dynamic funding cost relief and signalling e¤ects. Banks with a high level of non-performing loans and a low capital share were the most responsive to the measures and banks with higher uptakes in the credit easing operations reduced lending rates more than banks not participating in the program. Non-standard measures also contributed to normalise the dynamics of lending rates to households. It is well known that the pass-through of standard policy measures to household lending rates is generally lower than that to non-…nancial corporations, and that non-competitive pricing features explain this di¤erence. Even though the e¤ect is smaller, lending rates to households also fell in response to non-standard measures and banks with a high level of non-performing loans and a low level of capital responded most.
Non-standard measures compressed banks’lending margins signi…cantly. When pricing fric- tions are present, monetary policy may alter lending margins (Gambacorta, 2008; Alessandri
and Nelson, 2015). Changes in lending margins, in turn, might change the returns from matu- rity transformation activities with adverse e¤ects on their pro…tability and their market value.
We …nd that the compression of lending margins is larger for banks with a low level of capital, a higher exposure to sovereign debt, and a higher share of non-performing loans. Thus, while non-standard measures decreased funding costs for a class of banks, they hampered a component of their pro…tability, potentially making the banking system vulnerable to shocks.
We quantify the implications of standard and non-standard policies on the output gap and in‡ation using a workhorse macroeconomic model. We show that, by December 2015 and absent non-standard measures, in‡ation would have been 0.6 percent lower and the output gap 0.5 percent higher than actually recorded. We also …nd that the dynamics of the output gap and in‡ation that non-standard policies generate would have been insigni…cantly di¤erent from those obtained with a counterfactual perfect pass-through.
The rest of the paper is organised as follows. Section 2 provides a discussion of the transmis- sion of non-standard policy measures relevant for our study. Section 3 describes the data set.
Section 4 presents the empirical methodology. Section 5 discusses the pass-through of standard measures and section 6 the pass-through of non-standard measures. Section 7 highlights the macroeconomic implications of our …ndings. Section 8 concludes.
2 Channels of unconventional monetary policy transmission
The literature analysing the mechanics of monetary policy transmission is extensive. Prior to the period we consider, monetary policy in developed countries was conducted primarily by manipulating the short-term interest rate, which is a determinant of consumption and investment decisions. Three transmission channels are typically emphasised: the expectation channel, the interest rate channel, and the credit (bank lending and borrower’s balance sheet) channel, see e.g. Bernanke and Gertler (1995), or Mishkin (1996). These channels are well understood and su¢ ciently agreed upon to require no further discussion. When analysing non-standard policies, a number of additional channels may matter. Here we focus on those which may be most relevant for the credit easing (CE) and quantitative easing (QE) measures.
CE measures were introduced to support private sector credit ‡ows and to counteract weak in‡ation dynamics in situations where traditional interest rate instruments cannot be used.
They alter the size and the composition of the asset side of the central bank balance sheet.
QE measures were also conceived to ease private sector borrowing conditions and to counteract de‡ation risks. Contrary to CE, they involve a pre-de…ned expansion of the liabilities side of the
central bank balance sheet and no explicit reference to the assets it holds. While CE and QE measures are di¤erent, the channels through which they may a¤ect the economy are similar.
Thecost relief channel is the mechanism through which non-standard measures are expected to improve banks’ re…nancing conditions. This channel is prominent for CE measures, which are designed to reduce banks’marginal cost for targeted lending activities. By allowing banks to replace market funding with central bank funding, the policy directly impacts the supply of bank bonds. By arbitrage, this scarcity of securities should translate into lower yields for all bank bonds, including those issued by intermediaries not participating to the programme. In turn, this should lead to lower lending interest rates to the private sector. QE measures are likely to produce similar e¤ects: if bank bonds are perceived to be close substitutes for the asset purchased by the central bank, increases in the prices of the targeted asset should lead to a fall in bank bond yields (and this should be accompanied by a fall in the lending price.
QE measures are expected to activate aportfolio rebalancing channel. Central bank purchases change the relative supply of marketable assets and their yields, especially if the liquidity the private sector receives is not a perfect substitute for the liquidity contained in the assets sold.
Sellers of …nancial assets may thus rebalance their portfolios by buying assets with characteristics similar to those sold to the central bank. By increasing the price of sovereign bonds, the policies make lending more attractive relative to other investment possibilities, therefore supporting an expansion of bank lending. CE measures could also produce similar adjustments. Since the amount that banks borrow is a multiple of their eligible lending, the extra liquidity can be used to purchase, e.g., government and private sector securities. Moreover, the repayment of maturing bonds by banks participating in the programme may trigger portfolio rebalancing by bond holders. The portfolio rebalancing channel has been found important in the US by Krishnamurthy and Vissing-Jorgensen (2011), in the UK by Joyce et al. (2011) and in the euro area by Altavilla et al. (2015).
Finally, non-standard measures are a way for central banks to signal to market participants the intended path of future policy rates. Thissignalling channel works through a credible com- mitment to keep interest rates low: if central banks raise interest rates, they will su¤er losses on the assets purchased under the programme (Eggertsson and Woodford, 2003). The commitment mechanism may have two e¤ects: it may trigger a downward revision of expectations for future short-term rates - for CE this happens because of the long period of abundant liquidity implied by the maturity pro…le of the measures. It may also re-anchor in‡ation expectations, by pro- viding market participants information about future short-term interest rates. The signalling channel is expected to a¤ect the entire yield curve but the e¤ects depend on the maturity of
the bonds purchased. The importance of this channel has been studied in Krishnamurthy and Vissing-Jorgensen (2011), Bauer and Rudebusch (2014), and Altavilla et al. (2015).
3 The data set
The analysis makes use of two proprietary bank level data sets, regularly updated at the ECB.
The …rst, called Individual Monetary and Financial Institution Interest Rates or IMIR, contains information on individual deposits and lending rates charged by banks for di¤erent maturities and for di¤erent loan sizes. We construct bank-level composite indicators of borrowing rates of non-…nancial corporations and of households, weighting lending rates for di¤erent maturities and di¤erent loan size by new loans volumes. We do the same for deposit rates. Using new volumes at each maturity is important to make sure that the composite indicator correctly re‡ects the average lending rate applied by each bank. The weighting scheme we employ takes into account the fact that loans may be issued at ‡oating or …xed interest rate. The distinction is not crucial for lending rates to non-…nancial corporations - the vast majority of loan contracts are agreed at a ‡exible rate. However, it is relevant for lending rates to households. For example, in Italy, Ireland, Austria, Finland, Portugal the share of …xed rate loans for house purchases is below 25%; in Belgium, Germany and France, on the other hand, more than 80% of mortgage agreements have a …xed rate. Similarly, while savings banks tend to prefer loans at …xed rates, commercial banks have a more balanced portfolio of …xed and ‡exible mortgage loans. Clearly, these di¤erences matter: the higher the share of …xed rate loans is, the slower changes in the policy rate will be transmitted to household lending rates. We checked that the constructed indicators are representative of the banks’portfolios and made sure that no compositional biases resulted from to the fact that some banks lend primarily long term and others short term, or that some lend to small …rms (small amounts) and other to large corporations (large amounts).
From another proprietary data set called Individual Balance Sheet Indicators, or IBSI, re- porting the main asset and liability items of over 260 banks resident in the euro area from July 2007 to December 2015, we obtain data on the amount of outstanding loans, the exposure to sovereign debt and other relevant bank balance sheet information. We restrict the sample to head institutions and subsidiaries, so that each bank can be treated as independent (legal) entity.
This would not have been possible if branches were also included since a head institution must cover branch losses, with consequences for regulatory constraints and risk-taking behaviour.
The two data sets have rich cross-sectional information: we know whether a unit is a head institution or a (domestic or foreign) subsidiary; whether it is a large or a regional/local bank
and the country where it is located; and there are indicators of its business model (whether it lends more to non-…nancial corporations or to households, whether it funds itself more through capital or wholesale markets, etc.). The sample is representative of aggregate and of cross- sectional patterns: it covers about 75% of country aggregates and the cross-country distribution re‡ects the concentration in the area - see table A.1. For comparison, the stress testing exercise conducted by the European Banking Authority in 2014 included about 100 banks and balance sheet characteristics were only irregularly observed.
We match the information present in these data sets with information on bank bond yields (obtained from Markit - Iboxx), on regulatory capital ratios, and on non-performing loans (ob- tained from the commercial bank data provider SNL Financial) and on CDS (from Datastream).
Table 1 has information on the main items of banks’balance sheet.
The average lending rate to non-…nancial corporations (NFC) for banks belonging to the upper quartile of the distribution is about 130 basis points higher than the average lending rate for banks belonging to the lower quartile. This di¤erence increases to 160 basis points when we consider lending rates to households (HH). A similarly large di¤erential is present in bank bond yields (190 basis points) and in deposit rates (110 basis points). The spread in the distribution of sovereign debt exposure, capital ratios, leverage, non-performing loans, and credit risk is also quite large. We will use these balance sheet variables to group the cross-sectional distribution of monetary policy pass-through we obtain. Notice also that banks in our sample are medium sized, as measured by total assets, but very large banks are also present.
4 The econometric methodology
To address the questions of interest we use a cross-sectional Vector Autoregressive (VAR) methodology. In contrast to static pass-through equations, which are typically estimated with single-equation panel techniques, our approach has two main advantages. First, it allows for endogenous interaction between lending and funding conditions within a bank in response to monetary policy changes, an interaction that is neglected with single equation methodologies.
Second, it permits dynamic feedbacks between lending and funding conditions. These dynamic repercussions are disregarded in static models and improperly measured in single equation dy- namic setups. Furthermore, given the heterogeneities we have highlighted, standard panel data techniques, which assume that the dynamics of the endogenous variables are homogeneous, are inappropriate. The panel VAR model, suggested by Canova and Ciccarelli (2009) is potentially useful in this situation. However, the sparse nature of the dynamic interactions between banks
makes the setup ine¢ cient. Thus, to analyse the pass-through of monetary policy decisions, we proceed in two steps. In the …rst we use the time series dimension of the data, bank by bank, to estimate the dynamic response of lending rates (margins) to policy disturbances. In a second step, we sort the distribution of pass-throughs we computed using bank-speci…c characteristics and measure the average di¤erence between the upper and lower quartiles of the distribution.
Since the degree of pass-through might depend on country-speci…c factors (such as unemploy- ment, sovereign risk, or expectations about economic activity) and on bank-speci…c character- istics (such as cost of funding, and balance sheet conditions), we condition on country-speci…c factors when studying the monetary pass-through as a function of bank-speci…c characteristics.
Two important points about our methodology should be stressed. First, banks do borrow and lend in the overnight market but, over a month, the positions are generally averaged out.
Because dynamic interactions across banks are negligible and static interactions are likely to be small, computing the pass-through bank by bank entails little loss of information. Second, our two-step methodology is equivalent to allowing the intercept, the slope and the variance of the empirical model to feature (non-linear) interaction terms with bank-speci…c characteristics. Sa et al. (2014) used this approach to study the dynamics of UK housing market.
For each bank, the dynamic interactions among the endogenous variables are assumed to follow an unrestricted VAR. The contemporaneous relationships have a block recursive struc- ture. The vector of bank variables yit, is related to country-speci…c variables xjt and area-wide variables zt= [z1t; z2t]0, wherez1t is the policy rate, according to:
yit=Aixjt+Bizt+uit (1)
In addition, country-speci…c variables respond to area-wide variables but not vice versa:
xjt = Cjzt+ejt (2)
zt = vt (3)
Since the VAR for each bank features di¤erent endogenous variables, we constrainvtso that the response of the policy rate to its own shocks is the same in each VAR. This amounts to make the policy rate weakly exogenous with respect to country-speci…c and bank speci…c variables.
Each VAR includes three bank-speci…c variables (the lending rate, the deposit rate, and bond yields, when available) and three country-speci…c variables (10-year sovereign bond yields, the expected default frequency of non-…nancial corporations, and the unemployment rate). The
latter two variables are employed to capture business cycle conditions; the sovereign yield is used because a large portion of bank assets are in the form of local sovereign debt. Benchmark yields on 10-year sovereign debt are from DataStream; the expected default probability of non-
…nancial corporations is from Moody’s and the unemployment rate is from Eurostat. Other country-speci…c variables could be included but the chosen variables capture well those events that may a¤ect bank pricing decisions. It is important to include the expected default frequency for each country to control for potential changes in the riskiness of banks’customers which may lead to variations in the distribution of lending rates distinct from those we are care.
To make our results comparable with those present in the literature and to facilitate the interpretation of the results, we let zt include only the EONIA rate, which we use as a policy instrument. Adding other area-wide variables, such as the in‡ation rate or the unemployment rate, to the zt vector will re…ne the informational content of EONIA surprises but does not change the main features of the results. The choice of EONIA as the policy rate requires a few words of explanation. The three main ECB policy rates - the rate on the marginal lending facility, the rate on the main re…nancing operations (MRO), and the rate on the deposit facility (DF) –are available but move discontinuously with discrete jumps when the Governing Council decides rate changes. By contrast, the EONIA rate evolves daily, closely tracks the MRO rate in periods of normal liquidity and the DF rate when liquidity is abundant, and does not have a ‡oor at zero. Hence, the EONIA rate is a good indicator of the degree of monetary accommodation the ECB tries to achieve, both in normal and abnormal times.
As mentioned, slope heterogeneities prevents us from jointly using the time series and the cross-sectional dimensions for estimation. In addition, we can not rely on standard classical as- ymptotic theory for inference since the time series dimension of the panel is moderate. To derive the exact small sample distribution of the quantities of interest we employ Bayesian methods.
Let ibe the vector of bank-speci…c coe¢ cients, = [ 01; : : : 0N];let u =diag[ u;1; : : : u;N]be the covariance matrix of the disturbances, and let = ( ; ):We use a Normal-Inverse Wishart prior for ;modi…ed to incorporate i) a Minnesota prior, so that the empirical model for each unit is shrunk toward a vector of random walks with drifts; ii) a “sum-of-coe¢ cients” prior, which restricts the sum of the AR coe¢ cients of each equation, see Doan, Litterman, and Sims (1984), and iii) a “dummy-initial-observation” prior, which accounts for potential non-stationarities in the data, see Sims and Zha (1998). The vector of hyper-parameters controlling the informa- tiveness of the prior, ; is treated as random, to account for the uncertainty surrounding the speci…cation of the prior, and the posterior of ( ; ) is jointly computed.
Denote by mi a draw from the marginal posterior of i. An estimate of the responses of
bank variables to a conventional monetary policy disturbance is:
ymit !mi (L)vti= 1;2; :::N (4) where !i(L) is a 3 1 vector for each i, and L denotes the lag operator. Letting yit1;m be the response of the lending rate at time t for bank i, the pass-through at horizon h is:
P Tih;m Ph
=0!1;mi Ph
=0 m ; h= 1;2; :::H (5)
where z1t= (L)vt:Similarly, lettingwit1;m=y1;mit yit2;m= im(L)vt;whereyit2;mis the response of the deposit rate, the pass-through to lending margins uses 1;mi for !1;mi in (5). The dis- tribution of pass-throughs for each h is obtained exploiting cross-sectional and individual bank parameter variations. If lending rates responses were homogenous, i.e. !mi = !m 8i, cross- sectionally averaging (5) gives the pass-through obtained pooling cross-sectional information.
This quantity, however, would be di¤erent from the one computed in single pass-through equa- tions, because the contemporaneous and lagged feedbacks from deposit rates and bank bond yields to lending rates and the dynamics of country speci…c variables are disregarded. When we analyse standard measures, we use data from July 2007 to April 2014. The pass-through of non-standard measures is obtained using the parameter estimates for this sample and a path for certain endogenous variables from May 2014 to December 2015, as described in section 6.
Although the sample includes 260 banks, the actual number of banks we employ is smaller.
This is because bank bond yields are available only for a subset of (mostly large) banks, and some balance sheet characteristics might either be available for a limited number of banks (see Table 1) or simply not available for at least of 40 consecutive months – a required selection criteria for a bank to be included in our sample, to ensure that we have a su¢ cient number of time periods to analyse. In the baseline exercise, we eliminate bank bond yields from the VAR and consider a larger sample of banks (N = 168). We also consider the smaller sample of banks for which bank bond yields are available (N = 76). As we discuss later, the main conclusions we obtain hold in both samples.
5 The pass-through of conventional measures
The stylised evidence discussed in the introduction suggests that lending rates of banks located in …nancially stressed and non-stressed countries behave di¤erently. For example, the median
of the distribution of lending rates comoves less with policy rate changes in the stressed than in non-stressed countries in the 2011-2014 period. To examine whether this is still true when we condition on a number of factors, we compute lending rate responses to policy surprises and sort the distribution of pass-through by type of countries. Panel A of Figure 2 shows the distribution of lending rate responses and of the pass-through for all banks (…rst column), for banks operating in stressed countries (second column) and in non-stressed countries (third column). Responses are constructed assuming a persistent 100 basis point decline in the policy rate.
The median instantaneous pass-through is 0.25 and the median medium run pass-through is about 0.6, much lower than what was estimated prior to 2007 (Hristov, et al. 2014). Thus, the interest rate channel of monetary policy has considerably weakened. The distribution of pass-through is very dispersed: after 24 months, the highest posterior 68 percent interval goes from about 0.3 to 1.0. However, the location of banks does not explain this dispersion: the median of the distribution of pass-through has similar evolution and magnitude in both types of countries, and the highest posterior 68 percent intervals overlap. How, then, does one reconcile Figure 2 and Figure 1? There are two reasons why the patterns di¤er. The evidence in Figure 1 is unconditional while Figure 2 is constructed conditional on a monetary policy shock. There could be other country-speci…c disturbances driving the dynamics of lending rates; …nancial and technological shocks are few candidates that come to mind. Country-speci…c factors are considered in Figure 2 but not in Figure 1. Thus, the two …gures are consistent if, e.g., di¤erent probabilities of default are responsible for di¤erent lending rates across countries.
A concern when analysing the impact of policy surprises is that results might be a¤ected by the presence of non-standard policy measures. For example, since October 2008, the ECB has been lending liquidity to banks through …xed-rate full-allotment auctions. Since the EONIA rate adjusted accordingly, its innovations re‡ected both standard and non-standard monetary policy measures (see Ciccarelli et al, 2016). To show that the presence of non-standard provisions in the sample is inconsequential for the results, we matched the data with con…dential information about the participation of banks in the two 3-year Very Long-term Re…nancing Operations (VLTROs) conducted on the 20th of December 2011 and the 28th of February 20121and checked whether banks bidding in the operations (regardless of the amount actually taken up) displayed di¤erent pass-throughs than banks not participating the program. About half of the banks in our sample bid in one of the two operations, making the comparison statistically relevant. Panel B of Figure 2 shows that lending rate responses and the pass-through are similar in the two
1The …rst VLTRO providede489 billion to 523 banks, the second allottede530 billion to 800 banks. The net increase in liquidity provision was around EUR 500 billion.
groups of banks.
Figure 3 reports the mean value of the monetary pass-through in the upper and lower quar- tiles of the distribution and the 75 and 95 percent posterior intervals for the di¤erence. We group the distribution according to: (i) the exposure to domestic sovereign bonds, (ii) the (Tier 1) capital ratio, (iii) the degree of stability of the funding structure, as de…ned by Basel III, and (iv) the share of non-performing loans as a percentage of risk-weighted assets. To insure that the sorting variable is predetermined, we use bank characteristics recorded on July 2007.
Three features stand out. First, the pass-through is low for banks heavily exposed to domestic sovereign debt and with a weak capital position. For example, a persistent 100 basis point decline in the EONIA rate generates a medium run pass-through of about 110 basis points for highly capitalised banks, and of about 70 basis points for less well capitalised banks, and of over 120 basis points for the less exposed banks and of about 80 basis points for heavily exposed banks.
The fact that banks with a strong capital position and low sovereign bond exposure have a long-run pass-through greater than one may indicate that they strategically take advantage of their balance sheet position to expand their loan market share (see e.g. Gilchrist et al., 2015, for a model with such features). Our …nding that exposure to sovereign debt matters is consistent with the evidence in Drechsler et al. (2014), Altavilla et al. (2016) and Peydró et al. (2016) about risk-shifting incentives during crisis times. Note that the maximum di¤erence in average pass-through sorted by these two features is up to 40 basis points. Because the instantaneous pass-through is roughly similar, di¤erences across quartiles are due to the fact that poorly capitalised and highly exposed banks adjust their rates sluggishly over time. Van den Heuvel (2003) describes a model where lending rates of banks with weak capital position are sluggish because policy rate changes a¤ect bank capital. Our results suggest that the e¤ect on bank capital may be delayed, e.g., capital requirement may bind only on the dynamic adjustment path, but may be long lasting. The second outstanding feature is that quartile di¤erences in pass-through sorted by the stable funding variable are small and temporary. Finally, note that quartile di¤erences in pass-through sorted by non-performing loans are insigni…cant.
When the smaller sample of banks is considered, results are qualitatively similar when sov- ereign exposure and capital ratios are used to sort the distribution of pass-through (see Figure 4). Quantitatively, quartile di¤erences in pass-through sorted by sovereign debt exposure are larger - the maximum di¤erence is about 50 basis points. The stability of the funding structure is now irrelevant; the share of performing loans has some temporary, di¤erential e¤ect and banks with a high percentage of non-performing loans are slower in adjusting their lending rates for up to 8 months. Recall that this alternative sample includes primarily large banks - these are the
banks issuing bonds to fund their operations. Hence, the share of non-performing loans is an important factor in explaining lending rate di¤erences for large banks; stable funding di¤erences matter primarily for small banks.
In sum, the median monetary pass-through declined considerably between 2007 and 2014.
The decline is common to both stressed and non-stressed countries and consistent with the idea that frictions in pricing loans became, in general, more binding (see Gerali et al., 2010). Frictions interact with certain bank characteristics, in particular capital ratios and exposure to sovereign risk. This interaction seems responsible for both the disconnect between the dynamics of the policy rate and of the distribution of lending rates, and the large spread in the distribution of lending rates observed up to 2014. Banks became more prudent, charging customers at rates higher than one would have expected from the dynamics of the policy rate, because the deterioration of the asset side of their balance sheet and di¢ culties in securing appropriate funding threatened their long-run viability.
6 The pass-through of non-standard measures
In this section we evaluate whether the CE package –the targeted longer-term re…nancing opera- tions (TLTROs) announced in June 2014 –and the QE package –the asset purchase programme (APP) announced in January 2015 –helped to change the dynamics of the distribution of lend- ing rates and to reduce its dispersion. We also aim to identify the characteristics of the banks most a¤ected by central bank non-standard actions.
Although the EONIA rate responds to liquidity changes induced by the CE package, the VARs have no variable capturing central bank balance sheet expansion. To study the e¤ects of non-standard measures we thusproceed in two steps. We calculate the responses of certain variables to announcements of non-standard measures from May 2014 to December 2015 using a high-frequency event-study methodology (see e.g. Krishnamurthy and Vissing-Jorgensen 2011;
Altavilla et al. 2015). Then, we compare lending rates dynamics obtained mapping the policy- induced component of these variables onto individual bank lending rates with those obtained assuming that these variables evolved unconditionally since May 2014. We exploit the fact that CE and QE programmes alter the expected path of the EONIA rate - this is the signalling channel of section 2 - and a¤ect sovereign bond yields and banks credit risk, as re‡ected in the market price of bank debt - these re‡ect the portfolio rebalancing and cost relief channels.
This two-step approach is appealing because it captures the instantaneous …nancial markets e¤ects of non-standard measures, which are likely to be washed out with monthly data (see e.g.
Altavilla et al. 2016; Bluwstein and Canova, 2016; Ghysels et al., 2015). While non-standard measures may have been taken, in part, in response to failure of the lending rate distribution to adjust to policy rate changes, the methodology we employ is valid since, at the daily frequency, announcement dates are predetermined relative to the distribution of lending rates.
6.1 The impact of non-standard policies on …nancial markets
Quantifying the impact of non-standard measures on …nancial variables is challenging because many concurrent events a¤ected the EONIA rate and sovereign and bank bond prices in 2014 and 2015; expectations of US monetary policy tightening and oil price declines are two such events that come to mind. We use a narrative approach to identify policy surprises and run an auxiliary regression to obtain the path of the EONIA rate, of sovereign and of bank bond yields that would have occurred if the euro area had been hit only by non-standard policy announcements. The regression we run is
yt=aDt+bDt 1+cXt+et (6)
where Dt is a vector of dummy dates and Xt a vector of macroeconomic surprises in the euro area and the US and et iid disturbance. Dt includes twenty-eight announcement dates: for CE measures we include the Governing Council meeting dates held in May and June 2014; for QE measures we follow Altavilla et al. (2015) and use o¢ cial communications or hints about the likely implementation of the programme. The policy decision (made public on 3 December 2015) to extend the QE programme up to March 2017, to reinvest the principal payments, and to enlarge the set of eligible assets to regional and local government debt may have created anticipatory e¤ects. We take these into account by including the three Governing Council meet- ing dates since September 2015 and six dates associated with ECB’s o¢ cial speeches and data releases that led …nancial markets to revise their expectations about the likelihood of additional measures2. We do not consider anticipatory e¤ects of the 3 March 3 2016 announcement of the new targeted longer-term re…nancing operations programme (TLTRO-II), of the increase in the monthly purchases of the APP programme (from e60 to e80 billion) and of the enlargement of the list of eligible assets, because there were no hints that these measures were going to be
2The six events are Mr Draghi’s intervention in New York on 4 December 2015, which had clari…ed the easing potential of the December package; the speeches Mr Praet’s speeches on 22 September and 27 October 2015; the Bloomberg interview by Mr Constancio on the 25 November 2015; the market commentaries associated with the better-than-expected Economic Sentiment Indicator release on 29 October 2015; and the Reuters news regarding the growing consensus across Governing Council members on further deposit rate cuts on 9 November 2015.
implemented by December 2015. Xt is computed as the standardised di¤erence between the actual value of the data released and the consensus forecast made by professional forecasters – as collected by Bloomberg. The dependent variable ytmeasures the daily change in either the EONIA rate, or bank bond or sovereign bond yields. In (6) we use a two-day announcement window to allow for a sluggish market reaction to the news, which could have been possible given the novelty of the programmes. The results with a one-day window are similar. The policy component of the changes in the three variables is retrieved cumulating, monthly, the predicted path over the projection sample.
Figure 5 reports the median e¤ect (solid blue line) and either the cross-country or cross-bank variations (red dashed lines). The path for the EONIA rate is the same for all banks; the path of sovereign yields is country-speci…c, and the path of bank bond yields is bank-speci…c. Banks located in stressed countries bene…ted most from the announcements. By December 2015, their funding costs fell by 50 basis points and the sovereign bond rates fell by 100 basis points in the median. By comparison, a typical bank in non-stressed country saw funding cost and sovereign bond yields reductions of 30 and 60 basis points, respectively, in the median.
6.2 From …nancial variables to lending rates
To measure the e¤ects non-standard measures have on lending rates, we forecast them from May 2014 through December 2015, i) conditional on the path of the EONIA rate, of sovereign and of bank bond yields that would have occurred if the euro area were hit only by non-standard policy announcements, and ii) unconditionally. Formally, we compute:
uit+h =E yit+h1 j t; z1t+h; yjt+h E yit+h1 j t; z1t+h; yit+h (7) where t is the state of the economy at time t,yit+h1 is the path of the lending rate of bank i at horizonh= 1;2; ::::; z1t+h and yit+hare the policy-induced paths and z1t+h; yit+h the uncon- ditional paths for the three relevant variables. uit+h measures the response of lending rate of bank i to non-standard policy surprises (see Canova, 2007). The cross-sectional distribution of uit+h and for the implicit pass-through are in Figure 6.
Non-standard measures signi…cantly lowered lending rates - the median reduction is about 40 basis points - and the reduction is larger for banks operating in stressed countries (50 vs.
30 basis points). The implicit pass-through is also large: it is close to one in the median for stressed countries and about 0.8 for non-stressed countries by December 2015. Thus, the funding cost relief that banks enjoyed following were transferred to borrowers more in countries where
the monetary accommodation would have been most welcome. One may wonder whether the analysis is appropriate for banks located in, say, Greece or Cyprus, given that they were unable to obtain liquidity from the Central bank. While market segmentation is a concern in theory, it is irrelevant in our exercises because Greek and Cypriot banks are not in either of our samples - data on their non-performing loans is not available.
Balance sheet characteristics matter when it comes to explaining the reduction in the spread of the distribution of lending rate responses (see Figures 7 and 8). Non-standard measures were particularly e¤ective in lowering lending rates for banks with a high share of non-performing loans and low capital. The median di¤erence between the upper and lower quartiles of the distribution sorted by these characteristics is up to 40 basis points and di¤erences become strongly signi…cant after about 18 months. Di¤erences between the upper and lower quartiles of the distribution sorted by stable funding are insigni…cant. Some signi…cant di¤erence occurs when sovereign exposure is used as a sorting device, but the e¤ect is temporary and signi…cant only in the sample of large banks.
6.3 A few additional exercises
The analysis so far has been concerned with lending rates to non-…nancial corporations. Since we also have information about lending rates to households, we repeat the pass-through exercises with this variable and two goals in mind: we want to see whether the e¤ects of non-standard measures are robust; we are curious as to whether banks strategically used funding costs reliefs in the two markets to acquire market shares. Figures A.1-A.4 in Appendix A presents the pass- through distribution by type of country and by bank characteristics following conventional and non-standard policy surprises Qualitatively speaking, all the conclusions we obtained also hold for lending rates to households. In particular, in response to conventional policy changes the median pass-through is low (about 0.5) and the cross-sectional distribution is wide; the location of the bank does not explain the dispersion of the distribution of pass-throughs but the capi- tal ratio and the exposure to sovereign risk do. Non-standard measures increased the median pass-through; they were more e¤ective on the household lending rates in stressed countries, and a¤ected most banks with a high share of non-performing loans and with low capital. Quantita- tively speaking, the cross-sectional dispersion of pass-throughs in response to both standard and non-standard measures is smaller than for lending rates to …rms and the e¤ects of non-standard measures is also smaller.
It is of interest to evaluate whether bank credit risk is important in explaining the pattern
of lending rate responses to non-standard measures. It is well know that non-standard measures may also work through the so-called con…dence channel (see e.g., Bluwstein and Canova, 2016).
Here we want to evaluate whether banks with high perceived credit risk behaved di¤erently than banks with low perceived risk. The …rst row of Figure A6 in Appendix A shows that, in agreement with work by Acharya et al. (2015), banks that …nancial markets perceived to be more risky (as measured by individual banks CDS) were those whose lending rates to …nancial corporations fell most in response to the non-standard measures. The di¤erence with banks with low CDS was about 20 basis points, 18 months after the beginning of the programmes.
An additional piece of evidence supporting the idea that non-standard measures helped comes from the lending rates of banks participating in at least one TLTRO operation. While part of the decline in lending rates o¤ered by these banks is due to the fact that borrowers scaled back their recourse to wholesale funding, the second row of Figure A6 suggests that these banks lowered their rates substantially more than their non-participating peers. Interestingly, with lending rates to households this is no longer true: bidders and non-bidders display similar lending behaviour (see the last two rows of …gure A6). This accords with intuition since mortgages were excluded from the CE programme.
6.4 Side e¤ects? Pass-through to lending margins
There are several reasons to be concerned with the dynamics of lending margins – de…ned as the di¤erence between the lending rate to non-…nancial corporations and the deposit rate - in responses to non-standard measures. Several studies (e.g. Gambacorta, 2008; and Alessandri and Nelson, 2015) noticed that in the presence of frictions in pricing loans and deposits, changes in monetary policy may a¤ect the returns from maturity transformation activities and thus alter banks’pro…tability. In theory, the impact of non-standard measures on bank pro…tability is am- biguous. QE policies have two contrasting e¤ects: on the one hand, they ‡atten the yield curve, make maturity transformation less attractive, and thus hamper banks’ pro…tability. On the other hand, they may improve the capacity of borrowers to honour their commitments, increase the quality of the assets held in banks’ portfolio, and lead to a decline in provisioning needs.
Asset price increases also have a bene…cial impact on bank equity through valuation gains. In addition, as suggested by Drechsler et. al (2016), when the banking sector is imperfectly com- petitive, changes in monetary policy alter banks’e¤ective market power. Thus, when …nancial frictions matter, monetary policy in‡uences not only how much the banking system lends, but also how it is funded, the quantity of safe and liquid assets it produces, and its riskiness.
The dynamics of lending margins also matter from a di¤erent perspective. The magnitude of the pass-through is typically used to gauge the e¤ectiveness of the interest rate channel of monetary policy. However, the dynamics of deposit rates are equally important since they a¤ect the incentives of households to save. When the pass-through is imperfect, but deposit rates responses track lending rates responses, monetary policy may be as e¤ective as when the monetary pass-through is complete, but banks manipulate deposit rates to alter lending margins.
While an examination of the impact of monetary policy changes on banks pro…tability is beyond the scope of this paper, the dynamic responses of lending margins may give us some hints about the relevance of these concerns.
Non-standard surprises are associated with a substantial and statistically signi…cant com- pression of lending margins - about 20 basis points by December 2015 in the median (see Figure 9). The reduction is more pronounced for banks operating in stressed countries (30 vs. 15 basis points in the median), even though the cross-sectional variations in each group is large.
Banks with a low level of capital, with higher exposure to sovereign debt, and a higher share of non-performing loans have experienced a larger decline of their lending margins (see Figure 10). Thus, non-standard measures generate an important trade-o¤: mending the transmission channel of monetary policy and reducing borrowers’costs compresses lending margins of certain classes of banks. This trade-o¤ makes it important to consider the macroprudential consequences of non-standard measures, at least in the short run.
7 Macroeconomic implications
The monetary pass-through was quite imperfect for a large portion of our sample. Under work- ing capital constraints, this imperfection has implications for the evolution of the distribution of marginal costs that …rms face. With a standard speci…cations of …rms pricing decision, homo- geneous preferences and conventional technologies for producing monopolistically competitive goods, the distribution of marginal costs in turns imply through a cost channel (see e.g. Ravenna and Walsh, 2008), a distribution of good speci…c in‡ation rates.
To measure the e¤ects that the low pass-through obtained in the pre-2014 period had on in‡ation, we conduct a simple exercise using a standard New Keynesian model with sticky prices, habit persistence, and working capital described in appendix B, calibrating the parameters to match important features of the Euro area. We calculate the di¤erences in the in‡ation rate produced by the model with i) the actual paths of EONIA and of the median of the distribution of lending rates for the 2009-2014 period and ii) the actual path of the EONIA and a counterfactual
path for the lending rate, equal to the path of the EONIA rate plus a steady state spread. On average, the in‡ation rate should have been 52 basis points higher with the pass-through observed in the data. Core CPI average in‡ation rate in these …ve years in the euro area was roughly 1.0 percent. Thus, actual in‡ation rates were much lower than the reference value (2.0 percent) and the rate expected by the model. Three reasons may account for this outcome. First, in the period we analyze the slope of the Phillips curve has considerably ‡attened. Second, as Gilchrist et al. (2015) have suggested, …rms may trade-o¤ price and market share variations when they face …nancial frictions. In particular, …rms with lower marginal costs may slash prices more than expected by a standard Phillips curve to acquire market shares and thus increase revenues and pro…ts. If the goods produced by these …rms are dominant in core CPI in‡ation, the average behavior of in‡ation for the 2009-2014 period could be explained. Third, the in‡ation e¤ects of sticky lending rate dynamics were swamped by the contractionary e¤ects other shocks (e.g.,
…nancial shocks) had on aggregate demand.
To measure the e¤ects that non-standard policies had on the output gap and in‡ation, we compare two scenarios. In the …rst, the policy rate, the lending rate, and the deposit rates reproduce the paths the EONIA rate and the median values of the lending rate and the deposit rate distribution followed in the period 2014q2-2015q4 in response to the policy announcements.
In the second, the three rates are held constant at their 2014q1 values. In other words, we compute the di¤erence made by non-standard policies relative to the no-policy-change situation.
Figure 11 reports the paths for the output gap and in‡ation under the two policies. Non-standard policies had a positive and signi…cant e¤ect on in‡ation: the maximum impact is two quarters after the beginning of the programme and by 2015q4 the di¤erence with the no-policy scenario stabilises at around 0.6 percent. They also had a positive e¤ect on the output gap: by 2015q4 the output gap would have been lower by about 0.5 percent in the no-policy scenario. The mechanism inducing these changes is simple: non-standard policies, by decreasing the lending rate, decrease marginal costs for …rms (borrowing costs are lower) and this expands the aggregate supply, with positive e¤ects on employment. On the other hand, the fall in the deposit rate, increases the incentives of consumers to spend. With the parameterisation we employ, the aggregate demand e¤ect is large and drives in‡ation up and the output gap down.
Finally, we examine what would have happened to the output gap and in‡ation if non- standard measures were perfectly passed-through to lending and deposit rates. We construct a scenario where the lending and deposit rates perfectly comove with the policy rate and di¤er by average spreads observed prior to 2007, and compare it with the policy scenario constructed in Figure 11. Figure 12 presents the paths of the output gap and in‡ation under imperfect
and perfect pass-through. In‡ation would increase more in the short run under perfect pass- through, but by 2015q4 the e¤ect would be similar. The path for the output gap in the two scenarios is statistically insigni…cant. These results agree with Figure 6: the pass-through for the 2014q2-2015q4 period was already quite high.
A few important caveats need to mentioned. First, since the model features a representative agent, the redistributive e¤ects that non-standard policies have on borrowers and lenders are not captured in our calculations. Second, the model is highly stylised and since banks’pricing decisions are not explicitly modelled, the e¤ects of having a more stable and less risky banking system, which could be a by-product of non-standard policies, are disregarded in our exercises.
Third, the model does not feature any …nancial frictions on the consumer or the producer side. Since non-standard policies may have reduced these frictions, we are unable to evaluate the consequences they may have on the relationship between banks, producers and consumers.
Thus, the numbers we report represent the lower bound for the output gap and in‡ation e¤ects that non-standard policies may have had in the euro area.
8 Conclusions
This paper investigated the reasons why the interest rate channel of monetary policy in the euro area had weakened considerably over the last 10 years and studied how non-standard measures may have helped to mend the link between monetary policy and real activity. The analysis makes use of a novel and large data set covering European banks and employs information about their balance sheet characteristics and their funding structure. We exploit the time series dimension of the data, bank by bank, to construct the cross-sectional distribution of pass-throughs, use balance sheet characteristics to sort them, and measure the average di¤erence between the top and bottom quartiles of the distribution.
When considering standard policy rate surprises, we …nd a signi…cant fall in the median pass-through relative to the pre-2007 period. Balance sheet characteristics, in particular the capital position and the exposure to sovereign risk, explain the dispersion in the cross-sectional distribution of pass-throughs. The location of a bank, however, does not. Following a monetary expansion, poorly capitalised and highly exposed banks reduced their lending rates less than other banks because the deterioration of the asset side of their balance sheet and the di¢ culties in securing funding threatened their long run viability.
To evaluate the contribution of non-standard measures to the normalisation of lending mar- ket conditions, we isolate the impact these measures have on …nancial variables via an event-
study methodology. We then compare the lending rate dynamics obtained by mapping the policy-induced component of the EONIA rate, of sovereign yields and of bank bond yields onto individual bank lending rates, with those obtained assuming that these variables evolved uncon- ditionally since May 2014. The response of lending rates to non-standard measures was strong, the cross-sectional dispersion of responses was smaller than in previous years, and a larger pass- through was achieved. Conditions improved because of funding costs reliefs, dynamic portfolio rebalancing, and signalling e¤ects. Banks with a high level of non-performing loans and low capital were the most responsive to the measures. Large and small banks responded similarly to the measures and lending rates to both non-…nancial corporations and to households were equally a¤ected. Non-standard measures also produced a signi…cant compression of lending margins. Banks with a low level of capital, greater exposure to sovereign bonds and a high level of non-performing loans experienced the largest decline.
We quantify the macroeconomic impact of non-standard measures. We …nd that, by De- cember 2015 and absent non-standard measures, in‡ation would have been 0.6 percent lower and the output gap 0.5 percent higher than actually recorded and that, if the pass-through was perfect, the dynamics of the output gap and in‡ation would have been insigni…cantly di¤erent.
There are many important issues we did not address in the paper for reasons of space.
For example, whether the quality of loans improved after the implementation of non-standard measures, whether small …rms bene…tted from the improved lending conditions as much as large ones, and whether the maturity of the loans matters for the pass-through. In general, investigating the e¤ects of non-standard measures on the quantity and quality of loans would complement the pricing analysis of this paper. As mentioned, a study of the e¤ects of monetary policy on bank pro…tability is relevant from a macroprudential point of view, both for standard measures and for those non-standard measures which are becoming the norm in the developed world. An investigation of the bank external …nance premium, de…ned as the di¤erence between the cost of issuing bonds and the cost of …nancing the operations in the interbank market, could also be very useful to understand whether models of the …nancial accelerator apply to banks facing collateral constraints. Such an analysis could provide a further link between this paper and the literature studying …nancial constraints in macroeconomic models. The dynamics of lending rates may be driven by numerous shocks. Characterising whether lending rates are pro or countercyclical in response to these shocks may help us to select among various speci…cations of …nancial frictions proposed in the literature. We leave these issues for future research.
Acknowledgements
We would like to thank Viral Acharya, Fernando Alvarez, Giacomo Carboni, Mihnea Con- stantinescu, Domenico Giannone, Gabriel Perez-Quiros, Omar Rachedi, Silvia Miranda-Agrippino, Costanza Rodriguez D’Acri, Luca Sala, Ad van Riet and the participants of the Conferences:
Credit dynamics and the macroeconomy, London; the 2nd Oxford-NY Fed Monetary Eco- nomics Conference, Oxford; Unconventional Monetary Policy: E¤ectiveness and Risks, Rome;
the CAMP workshop on Commodity prices and monetary policy and of seminars at European Central Bank, Federal Reserve Board, Bank of Spain, Central Bank of Slovakia, Universidad de Murcia, Central Bank of Lithuania, Central Bank of Argentina, Central Bank of Chile, Central Bank of Israel, Central Bank of Belgium for comments and suggestions. Canova’s portion of the research was …nanced, in part, by the Spanish Ministry of Economy and Competitiveness, Grant ECO2015-68136-P and FEDER, UE. The views expressed in the paper are solely ours and do not necessarily re‡ect those of the European Central Bank or the Eurosystem.
References
[1] Acharya V. V., T. Eisert, C. Eu…nger, and C. Hirsch (2016) Whatever it takes: The Real E¤ects of Unconventional Monetary Policy", NYU manuscript.
[2] Acharya V. V., B. Imbierowicz, S. Ste¤en and D. Teichmann (2015). Does the Lack of Financial Stability Impair the Transmission of Monetary Policy?, NYU manuscript.
[3] Acharya, V. and S. Ste¤en (2015). The Greatest Carry Trade Ever? Understanding Euro- zone Bank Risks, Journal of Financial Economics,115, 215-236.
[4] Alessandri P. and B. Nelson, (2015). Simple Banking: Pro…tability and the Yield Curve, Journal of Money, Credit and Banking, 47(1), 143-175.
[5] Altavilla C., Carboni G. and R. Motto (2015). Asset Purchase Programmes and Financial Markets: lessons from the Euro Area, European Central Bank working paper 1864.
[6] Altavilla, C, Giannone D., and M. Lenza (2016). The …nancial and macroeconomic e¤ects of OMT announcements, International Journal of Central Banking, 12, 29-57
[7] Altavilla C., Pagano M., and S. Simonelli (2016). Bank Exposures and Sovereign Stress Transmission, CEPR Discussion Paper 11269.
[8] Bauer M. D. and G. D. Rudebusch, 2014. The Signaling Channel for Federal Reserve Bond Purchases, International Journal of Central Banking, 10(3), 233-289.
[9] Bernanke B., and M. Gertler (1995). Inside the black box: The credit channel of monetary policy transmission. The Journal of Economic Perspectives, 9, 27-48.
[10] Bluwstein, K. and F. Canova (2016). Beggar-thy-neighbour? The international e¤ects of ECB unconventional monetary policy,International Journal of Central Banking, 12, 69-120.
[11] Canova, F. (2007) Methods for applied macroeconomic research, Princeton University Press, Princeton, NJ.
[12] Canova, F. and M. Ciccarelli (2009). Estimating multi-country VAR models,International Economic Review, 50(3), 929-961.
[13] Ciccarelli, M., A. Maddaloni, and J. L. Peydró (2013). Heterogeneous transmission mecha- nism: monetary policy and …nancial fragility in the eurozone.Economic Policy, 28, 459-512.
[14] Ciccarelli, M. A. Maddaloni, and J.L. Peydró (2016). Trusting the bankers: a new look at the credit channel of monetary policy transmission. Review of Economic Dynamics, 18, 979-1002.
[15] De Santis, R. and P. Surico (2013). Bank lending and monetary transmission in the euro area.Economic Policy, 28, 423-457.
[16] Doan, T., Litterman, R. and C. Sims (1984). Forecasting and Conditional Projections using realistic Prior Distributions, Econometrics Reviews, 3, 1-100.
[17] Drechsler I., Drechsel T., D. Marques-Ibanez, and P. Schnabl (2014). Who Borrows from the Lender of Last Resort?, Journal of Finance, forthcoming
[18] Drechsler, I., A. Savov, and P. Schnabl, (2016). The deposit channel of monetary policy, NBER working paper 22152.
[19] Eggertsson, G. and M. Woodford (2003). The zero bound on interest rates and optimal monetary policy. Brookings papers on economic activity, 1, 139–233.
[20] Gambacorta L. (2008). How do banks set interest rates? European Economic Review, 52, 792-819.
[21] Gambacorta L. and P. E. Mistrulli (2014). Bank Heterogeneity and Interest Rate Setting:
What Lessons Have We Learned since Lehman Brothers? Journal of Money, Credit and Banking, 46, 753-778.
[22] Gerali, A., Neri, S., Sessa, L., and F. M. Signoretti (2010). Credit and banking in a DSGE model of the Euro Area. Journal of Money Credit and Banking, 42, 107-141.
[23] Gilchrist, S., Schoenle, R., Sim, J. and E. Zakrajsek (2015). In‡ation dynamics during the
…nancial crisis, Boston University, manuscript.
[24] Ghysels, E., Idier, J., Manganelli, S. and O. Vergote (2015). A High Frequency Assessment of the ECB Securities Markets Programme, forthcoming,Journal of the European Economic Association.
[25] Hristov, K., Hulsewig, O. and T. Wollmerhouser (2014). Interest rate pass-through in the Euro area during the Global Financial Crisis.Journal of Banking and Finance, 48, 104-119.
[26] Holton S. and C. Rodriguez d’Acri (2015) Jagged cli¤s and stumbling blocks: interest rate pass-through fragmentation during the Euro area crisis, ECB WP1850.
[27] Illes A., Lombardi M. and P. Mizen (2015). Why did bank lending rates diverge from policy rates after the …nancial crisis?, BIS Working Papers 486, Bank for International Settlements.
[28] Joyce M, A Lasaosa, M Tong and I Stevens (2011). The …nancial market impact of quan- titative easing in the United Kingdom, International Journal of Central Banking, 7(3), 113-61
[29] Kashyap, A. and J.C. Stein, (1995). The impact of monetary policy on bank balance sheets.
Carnegie Rochester Conference Series on Public Policy, 42, 151-195.
[30] Kashyap, A., and J.C. Stein, (2000). What do a million observations on banks say about the transmission of monetary policy. American Economic Review,90 (3), 407-428.
[31] Kishan R. and T. Opiela (2000). Bank size, bank capital, and the bank lending channel.
Journal of Money, Credit and Banking, 32, 121-141.
[32] Krishnamurthy, A. and Vissing-Jorgensen, A. (2011). The e¤ects of quantitative easing on interest rates: channels and implications for policy.Brookings Papers on Economic Activity, 43(2), 215–287.
[33] Mishkin, F. (1996). The channels of monetary transmission: Lessons for monetary policy.
NBER working paper no. 5464.
[34] Peek, J. & Rosengren, E.S. (1995). Bank Lending and the Transmission of Monetary Policy, in Peek J. and Rosengren E.S. (eds.), Is Bank Lending Important for the Transmission of Monetary Policy?, Federal Reserve Bank of Boston Conference Series, 39, 47-68.
[35] Peydró J.L., A. Polo, and E. Sette (2016) Monetary Policy at Work: Security and Credit Application Registers Evidence, UPF manuscript.
[36] Ravenna, F. and C. Walsh (2008). Optimal monetary policy with the cost channel,Journal of Monetary Economics, 53(2), 199-216.
[37] Sa, F., P. Towbin and T. Wieladek (2014) Capital in‡ows, …nancial structure and housing booms, Journal of the European Economic Association, 12(2), 522-546.
[38] Sims, C. and T. Zha (1998). Bayesian methods for multivariate models,International Eco- nomic Review, 39, 949-968.
[39] Stein J.C. (1998), An Adverse-Selection Model of Bank Asset and Liability Management with Implications for the Transmission of Monetary Policy, RAND Journal of Economics, 29(3), 466-486.
[40] van den Heuvel, S. (2002). Does bank capital matter for monetary transmission? Federal Reserve Bank of New York Economic Policy Review, May, 259-263.
[41] von Borstel J., Eickmeier, S. and L.Krippner (2015). The interest rate pass-through in the euro area during the sovereign debt crisis, Bundesbank manuscript.
